[OpenAFS-devel] Re: fileserver -> client NAT ping

[Andrew Deason]

On Wed, 14 Aug 2013 23:01:23 +0200 (CEST)
Harald Barth <haba@kth.se> wrote:

> > The addresses in use are not a hint that there is a problem.
> 
> I say it's a hint, but we might want a better hint.
> 
> Can we somehow follow uuids when ports are changing because of NAT? I
> mean after the first timeout, so we don't need to timeout again.

I mentioned in some other email that it's possible to have more robust
'detection', but I didn't think it was worth the effort. Trying to make
such decisions by tracking behavior like that gets a little complex for
the host package (which is already doing simpler things by comparison
IMO, and is already quite complex).

But yes, I think to some degree it's possible to detect NAT 'breakage'
like that. But it'll never be 100% accurate.

> The question is how to figure out if this actually makes things
> better, but we won't know if we don't try. I volunteer for testing
> such a patch, but I'd like to have a way of logging  

Do you have reasons to believe you would be affected by clients with bad
NATs? Testing this I think isn't really useful under lab/artificial
conditions; I mean, we generally know what will happen given specific
conditions. So it's only really useful with 'real' clients and such.

I have some code to do this; the tip of it is here
<http://gerrit.openafs.org/#change,10147> /
72f90085503afafd1a5b7d510de0cf700b693865. You can either use that tree
directly or apply the patches from gerrit 10144-10147.

That implementation probably isn't what I'd want for an actual release,
but it should be functionally equivalent to what I'd want to do. (Just,
the interfaces and such I used are obviously just done quickly to test
this out)

-- 
Andrew Deason
adeason@sinenomine.net